Code for Enhancing Generalization in Sketch-Based Image Retrieval through Single and Multi-Source Domain Adaptation

Huang, Mengqing (2025) Code for Enhancing Generalization in Sketch-Based Image Retrieval through Single and Multi-Source Domain Adaptation.

This study is conducted on the CIFAR-100 dataset, with only 50 randomly selected object classes used for training. The remaining classes are reserved for zero-shot testing. Each pre-trained model is fine-tuned on CIFAR-100 to adapt itself to this task. Subsequently, various model variants are created by adding a single linear probe layer for subsequent zero-shot performance evaluation. These models are then tested, collecting metrics such as error rate and kappa. Finally, the trade-off point between these metrics is calculated and visualized in graphs, ready for analysis.